OpenAI Slashes GPT-5.6 Luna Price by 80% Amid Intense Open-Weight Competition

The artificial intelligence industry is experiencing a rapid commoditization of raw reasoning power. In a move that signals a fundamental shift in the market, OpenAI has slashed the price of its lightweight GPT-5.6 Luna model by 80 percent.

The dramatic price reduction arrives just three weeks after the GPT-5.6 family was introduced to the public. Historically, enterprise pricing tiers are set carefully and revisited annually, making this an unusually aggressive timeline for a frontier model line.[2]

The new pricing structure dramatically lowers the financial barrier to entry for developers building AI applications. GPT-5.6 Luna now costs $0.20 per million input tokens and $1.20 per million output tokens, down from $1 and $6 respectively.

Meanwhile, the mid-tier GPT-5.6 Terra received a 20 percent price cut, dropping to $2 per million input tokens and $12 per million output tokens.[1]

The flagship model, GPT-5.6 Sol, maintained its standard pricing but gained a new "Fast mode." This premium option delivers up to 2.5 times the processing throughput for twice the cost, without sacrificing the model's underlying intelligence.[1]

The primary catalyst for this sudden repricing is intense pressure from open-weight models, particularly those emerging from China's rapidly advancing tech sector.

Companies like Moonshot AI, with its Kimi K3 model, and Zhipu AI, with GLM-5.2, have been aggressively courting international developers with highly capable, low-cost alternatives.

Open-weight models—systems where the underlying trained parameters are publicly accessible—allow developers to deploy AI on their own infrastructure. This accessibility has fundamentally changed enterprise buying habits, eroding the lock-in power of proprietary labs.

Instead of defaulting to the most famous proprietary model, engineering teams are increasingly cost-conscious. They are utilizing routing software to automatically send workloads to the cheapest model capable of completing a specific task.[2]

To maintain its market dominance, OpenAI had to address the "cost per task" metric. While some open-weight models boast lower per-token prices, they often require more tokens to arrive at the correct answer, making the net cost roughly equal.

By dropping Luna's price by 80 percent, OpenAI has instantly propelled the model into the "most attractive" tier in intelligence-per-dollar rankings, placing it above Chinese rivals like MiniMax's M3.

The mechanism behind this massive price cut is perhaps the most compelling part of the story: recursive self-improvement.

OpenAI reportedly used its most powerful model, GPT-5.6 Sol, to analyze and rewrite the GPU kernels that run the smaller Luna model.

By identifying production inefficiencies and optimizing techniques like speculative decoding—where a smaller model drafts text and a larger one verifies it—the system effectively made itself cheaper to operate.

This self-optimization marks a critical milestone for the industry. When AI models can reduce their own compute overhead by rewriting their own infrastructure code, the cost floor for intelligence drops exponentially.

For everyday users and enterprise developers, this price war unlocks new possibilities, particularly for "agentic workflows." In an agentic system, an AI doesn't just answer a single prompt; it loops continuously, researching, writing, and correcting its own work until a complex goal is met.

These autonomous loops consume massive amounts of tokens. At previous prices, running thousands of agents was financially prohibitive for most startups. At 20 cents per million input tokens, developers can now afford to let models "think" longer and iterate more frequently.

Ultimately, the foundation model market is beginning to resemble traditional utility infrastructure. As raw intelligence becomes abundant and cheap, the competitive advantage is shifting from who has the smartest model to who can operate at the most efficient global scale.